Strategy of European local municipalities officials to deal with financial decline in the context of the Covid-19 pandemic: An analysis of the influence of risk-taking and debt preferences by survey experiment

(1)

Strategy of European local municipalities officials to deal with

financial decline in the context of the Covid-19 pandemic:

An analysis of the influence of risk-taking and debt

preferences by survey experiment.

Master thesis

Author: Guillaume Nedelec

MSc Public Administration - specialisation Economics & Governance

Year: 2020-2021

(2)

Abstract

This study analyses the relationship between negative financial information and preference for a directing state and the influence of risk-taking and debt preference moderators in the context of Covid-19 pandemic. We aim at observing the empirical effect of providing a group with a treatment containing negative financial information on preference for increased public services provision by asking council members from England, France, Netherlands, Spain and Switzerland using a survey-experiment method. We first embed our study in the frame of globalisation from 1970’s to 2008 financial and economic crisis to contextualize growing austerity measures taken across Europe afterwards. From the end of 2019, coronavirus crisis became the centre of attention of citizens, voters, policymakers, decision-makers and politicians going beyond the boundaries of a health crisis towards a new incoming economic crisis. Using insights from behavioural public administration, at the intersection between public administration and psychology, and working with randomized survey-experiment, we argue they are of primary relevance to understand causal relationship to understand how treating a group with negative financial information could result in difference of council members directing state preference.

After having supported and argued towards relationship between independent, dependent and moderating variables on theoretical grounds, we tested those effects empirically.

By performing manipulation checks, we made sure respondents were equally split between the control and treatment groups regarding their individual characteristics. Descriptive statistics and regression analyses revealed the existence of significant effects between variables of interest. Those results constitute a solid basis for interpretation and analyses in the frame of our study meaning contributions to theory, practice and research. Also, we discuss the relevance of those results beyond internal validity of the model we developed meaning limitations and suggestions for future research.

Keywords

Directing state; negative financial information; local government; randomized survey-experiment; behavioural public administration.

(3)

Acknowledgments

Singularly, I would like to thank Mr. van der Voet, my supervisor for giving me a chance to participate in his capstone project and collaborate with him on some parts of the study. Thanks to Mr. van der Voet, I was able to write this thesis benefiting from his expertise, trust, support and benevolence, always in a way that pushes me to improve and do better.

Also, I would like to thank my fellow colleagues who participated with me to the capstone project: Rafael Diogo Carrilho, Marine Marty and Timo van der Kraan. Without them, I would not have been able to collect answers from Spain, Switzerland and England but also the Netherlands thanks to Mr. van der Voet.

Finally, for his course about political economy in international perspective, by teaching me key concepts of globalisation, I would like to thank Mr. Olaf van Vliet.

(4)

1.Introduction

From 1970s onwards, policymakers and politicians have to solve a complex equation: doing

more with less. We argue three important waves are relevant in tracing welfare state resource scarcity: post-war globalisation, financial and economic crisis in 2007 and coronavirus pandemic in 2020. We will explain then the relevance of our study embedded in this broader scope and value added of the research.

1.1. Post-war globalisation

From 1970s and 1980s, globalisation involved economies openness, increase of trade as part of GDP but also an eased mobility of goods, capital, people and information. It appeared high-skilled individuals could benefit from globalisation by accessing advantageous work opportunities in another country, improving their welfare. In spite of this, it also demonstrated a weakening of a considerable part of nation-state population. Low skill labour is attracted in the most developed countries because of the lowest costs of labour for companies. Faced with competitive pressure exerted by labour workers coming from abroad moves, low-skill national workers are even more exposed to unemployment, poverty, income inequality and other underlying socioeconomic issues (Walter, 2010).Politicians had to deal with less financial resources while citizens were asking for more social protection against increased risks induced by globalisation and public intervention. By expanding the welfare state it refers to the compensation hypothesis, in opposition to the efficiency hypothesis aiming at retrenchment. However, theoretical concerns animated debates in the field of political economy. Scholars have argued and highlighted the distinction between welfare state retrenchment (e.g. a modification of the state structure in terms of budget allocated to a specific sector like health or education) and cutbacks (e.g. a reduction in public spending in any, multiple sectors or in general) as necessary (Green-Pedersen, 2004).Though liberalisation and privatisation operated in the welfare-state led by prominent proponents such as Margaret Thatcher, former UK Prime-minister or Ronald Reagan, former US President could have figured a retrenched welfare-state, authors observed an expansion of the welfare state (Rodrik, 1998). Though scholars such as Iversen & Cusack (2000) highlighted the importance of internal pressures for state compensation and the limited effect of globalisation, they agreed on the fact state expanded.

1.2. Reform and austerity in the 2008 financial crisis aftermath

Over last decade, economic and financial crisis started in 2007 considerably impacted the room for public spending and investment. Despite the speculative bubble burst in the United States, it

(7)

spreads worldwide because of even more interconnected and interdependent economies and financial markets due globalisation (Eun, 1991). The crash originated from real estate and subprime system. In a few words, banks intensively loaned money to households that didn’t have enough good repayment capacities in comparison to the amount borrowed. The rationale beyond is that, even if individuals cannot reimburse anymore the credit, banks would benefit from mortgage as long as real estate market continues to rise. A considerable lasting inflation incentivises loaners towards more and more profit, taking uncountable risks. Bankers didn’t realise the market value of houses differed increasingly significantly from intrinsic value (e.g. determined objectively by using diverse indicators). Their vision was blurred by speculative bubble euphoria, operating a behavioural change leading to non-rationality. Bubble exploded when real estate market prices started to decrease. This world crisis converted a few years after in what have been called a “sovereign debt crisis”. EU governments increased their public debts to avoid banks collapse and bankrupts, emitting bonds on the market and borrow money at a very low interest-rate set by the European Central Bank (ECB). The latter one used an unconventional tool for monetary policy, quantitative easing, meaning the ECB particularly bought government bonds to increase money supply in the economy to ensure a sufficient level of inflation. Still, it resulted in less financial resources for the state and public institutions to implement public policies and programmes relying on austerity plans (Mehra, 2012).

1.3. Coronavirus health crisis and a new economic crisis shadow

At the end of November 2019, a new virus unknown from scientists to date has started to reproduce and spread over the population. Being a member of the coronaviruses category, it causes the disease covid-19. Originating from Chinese city of Wuhan, the way pandemic started and transmitted to human is still unclear (Kuznia & Griffin, 2020).Freedom of move, human flux and open frontiers, especially in the Schengen area, facilitated cross-borders contamination and import of the disease in a new, not yet impacted, state. To date of mid-April 2019, from China the covid-19 pandemic has been reported to have infected people in 185 countries causing the death of at least 161,000 humans (The Visual Data and Journalism Team, 2020). Nonetheless, it is impossible to have a full and set data as long as the outbreak doesn’t stop. More, we already know statistics provided suffer from underestimation bias associated with numerous reasons: some countries don’t allow a transparent access to data and are suspected of manipulating numbers, countries have different reporting methods, some countries perform less covid-19 testing so having less positive cases in rate than others, some countries have a low quality of healthcare and curing or inhabitants suffering from several diseases already then being incapable of determining whether an individual died of covid-19 or another infection, etc. In comparison with previous pandemics occurring centuries ago, what is sure is that globalisation facilitated the spread of the coronavirus at

(8)

world-level. Beyond the sanitary and health effects of the covid-19, economic effects are expected to be huge when looking at public interventions already engaged to consequences of lockdown and quarantine. As public places judged as non-vital are for almost all of them closed, welfare is threatened. Temporary unemployment has been decided by a lot of companies in the EU and even unemployment in the US that observed 22 million individuals applying for benefits in only four weeks (Tappe & Luhby, 2020). A country like France decided to grant different benefits for healthcare workers but also the poorest households to compensate those who are the more exposed to risks whatever their nature is: health or socioeconomic. If the degree, scope and extent of how people will be assisted vary consequently according to nation-states models, on overall governments engaged greater and often impressive public spending to prevent welfare losses. EU finance ministers reached an agreement for an emergency fund of 500 billions of euros, in the framework of the European Stability Mechanism and US President Donald Trump signed a 8.3 billions of US dollars to limit at maximum the negative effects of coronavirus on the economy (Boffey, 2020; Hirsch & Breuninger, 2020). Again, how harmful precautionary measures have been for the economy are still estimates and would need further investigation after the crisis. Yet, what financial resources will politicians have at their disposal raises uncertainty about their ability to maintain budgets under control without relying dramatically on public debt increases or significant cutbacks.

1.4. Focus of the study

In the 2008 financial crisis aftermath, Lodge and Hood (2012) studied Public Service Bargains (PSBs) within the Organization for Economic Cooperation and Development (OECD). They notably argued although facing similar pressures, states can opt for four different responses concerning public services provision at the local level. Especially, we choose to test the effect of negative financial information on the directing state approach (Lodge & Hood, 2012) implying greater intervention in the economy by public authorities. We chose to focus on directing state approach for two main reasons. First, as illustrated by Lodge & Hood (2012), former economic and financial crisis started last decade observed a comeback of a state having to intervene in the economy to prevent bankruptcy and collapse of national economy. Second, similarly, states already played a major role in managing coronavirus crisis by measures of partial unemployment, increased revenues by pay rises for public service workers or increase of social benefits allocated. Our aim when making this study is to base on robust conceptual and empirical findings to focus on local government official’s behavioural response and give valuable insights about their preferences for a directing state within coronavirus context. Jordan & Audia (2012) demonstrated negative financial information and low performance triggers behavioural response from decision-makers. They explain behavioural response is variable and different ways of reasoning for decision-making account for those differences among individuals However, we think two moderators are especially

(9)

relevant for our analysis. Our first moderator concerns risk-taking as an individual characteristic. Frey & all (2017) constructed interesting argumentation towards conceiving risk-taking as a fixed property stable over time. The extent to which individuals take risks influences policy preference because they involve different degrees of risks. Our second moderator is about individual preference for an increase in public debt. Including a measure for debt preference is a more relevant way of accounting for political preference and membership. Indeed, political parties are not the same among countries hence making it not feasible to have a common basis for all countries. Debt preference is consistent with choosing for a policy approach because it will influence which one individual prefer as different debt preference leads to different preference for a directing state. As authors demonstrated increased polarisation of politics within the European Union, preference for debt is a strategic way of constraining future action of government. Alesina & Tabellini (1990) and Persson & Svensson (1989) illustrated the concept of “strategic debt” leading to higher debt increase. Then, if we observe consistent levels of debt increase preference when asking respondents, we can expect them to be willing to constrain future government set of options. Whether they are willing to results in different preference for a directing state.

Thus, as a research question, we observe what is the effect of negative financial information municipal councillors’ preference for a directing state and how this effect is moderated by risk-assessment and debt increase preference, in the context of coronavirus pandemic.

1.5. Relevance of the study

In the specific context of coronavirus, conducting such an analysis is particularly interesting because the underlying economic crisis is different in nature. As states opted for different strategies to tackle the crisis, it is interesting to take into account municipal councillors answers from survey-experiments from five different countries all OECD members such as France, Netherlands, England, Spain and Switzerland. Because they vary in institutional features and strategies, we aim at observing how a single model is good in explaining overall variation, knowing that Lodge & Hood (2012) framework is also based on OECD observations. While the 2008 economic crisis was caused by the collapse of the financial system due to risky investments and strategies, economic crisis world is actually facing is the result of a virus that obliged governments to take harmful measures for the economy to protect health of citizens. In terms of psychology and attention, the coronavirus pandemic acted as a “focusing event” (Sabatier & Weible, 2014) at least and could provoke a paradigm shift (Kuhn, 1962). It would not be surprising to observe change in public policy priorities with a change in public agenda in response to a focusing event (Sabatier & Weible, 2014). More, such an event seriously challenged dominant view of globalization and liberalism with the economic crisis effects post-2008. Whether a paradigm shift occur or simply was it a call for reform in the financial and economic spheres have been debated in the literature (Pollitt, 2010; Mackintosh,

(10)

2015). For sure, the recession coming for the future years will put even more pressure on public finances and spending. In those changing priorities era, some individuals are willing to take more or less risks because the “window of opportunity” for agenda change has a limited duration (Sabatier & Weible, 2014).

Lodge and Hood (2012) article provides a solid theoretical framework about positions a state can adopt when facing resource scarcity. However, authors indicated their framework does not limit to state level and could also be applied to other government levels. Then, we decided to apply and test the policy framework provided by the authors at the level of municipal councillors.

By providing negative performance feedback about finances and spending to a treatment group by a survey-experiment, we expect a behavioural effect on respondents’ perceptions, then observing how this stimulus affect their preferences for a directing state. As we apply this concept at local level, we derived the subsequent notion of what it would imply for a local government. In our study, it is understood as an increased reliance on public services provision at municipal level, taking over and handling more tasks including economic policy. Then, individual decision-makers would be the key actors of public services management.

We observed little attention have been devoted to individual decision-makers level while lots of macroeconomic studies have been conducted after 2008 and the financial crisis. Some analyses concentrated on local government level but did not consider individual decision-making (Paulais, 2009; Council of Europe, 2010). Indeed, most of the papers addressed efficiency issues or how to improve local management. Also, coronavirus measures were mainly looked at state level whereas public service provision was ensured, and new measures implemented by officials elected in local government and municipalities. Our choice of moderating variables is based on individual characteristics about risk-taking and debt increase preference. Thus, we give more evidence individual factors can be useful in explaining causal relationship between negative financial information and preference for a directing state. We argue microlevel characteristics are of specific interest in our study because they are more likely to reveal observed preferences more closely to real preferences. Making such an analysis at local government unit of observation and not at individual level cannot illustrate those differences in people characteristics. Using survey-experiment is particularly relevant when being interested in individual traits because respondents are directly asked to provide an answer by themselves. When conducting such an analysis a macrolevel, some individuals might influence the final answer more than others then individuals have a different weight. Also, answers may be biased because having to express views or reveal preference in presence of others. Thus, they may give answers that would satisfy others or be influenced by them to not weaken their reputation in case of unpopular view. We would observe a considerable discrepancy between real values and observed values. Our analysis would be less accountable for explaining real-world phenomena. Conducting a survey experiment also

(11)

contributes to expand its use in the field of public administration by demonstrating it can produce significant results then being reliable. Also, it allows us to observe for moderating effects that cannot be observed or at least difficulty otherwise. It is very useful in exploring and understanding intentions and preferences.

Finally, few studies have been conducted in the coronavirus context. Social science precisely intents to explain what happens in real-world by relying on theories and concepts. Our study aims at revealing whether such a framework developed by Lodge & Hood (2012) is relevant in explaining variation in municipal councillors’ preference for greater public intervention in the economy, when applied in the specific coronavirus and underlying economic crises environment. It is about testing if this model can be used in different environments or another model would be better in the specific case we are making our analysis.

(12)

2. Theoretical background

This chapter aims at explaining the central concepts that constitute the foundations of the thesis. More, we will establish links and connections between those concepts and derive subsequent hypotheses.

2.1. Concepts and theory foundations

This chapter will review the concepts of performance feedback reaction, policy positions, risk-assessment and debt preferences.

2.1.1.

Performance feedback reaction

Our primary concern is to explain the concepts we use in our study. Our starting point then, is about negative financial information. We elaborated this notion on the basis of Jordan & Audia (2012). The authors provide core knowledge about why and how individual decision-makers react to negative performance feedback. First, they recall organisational theory has been mainly influenced by bounded rationality (Simon, 1955). Bounded rationality poses individuals as facing constraints making them unable to opt for perfectly optimal solutions when having to make a choice. Those boundaries to optimality can be a practical limitation such as time in the sense only 24 hours are available in a day and we cannot extend its duration. Another cognitive limitation can be imperfect information: when making a choice, individuals do not always have all information at their disposal or suffer from an information asymmetry, meaning they have less and incomplete information than other individuals.

Those examples depict situations where individuals still have to make a decision in an imperfect environment. However, behavioural economics is not a study field that caught the attention of many of public administration scientists (Jones, 2003). Yet, Simon (1947) highlighted the need for relying on psychology to study public administration decades ago. From bounded rationality (Simon, 1955), Cyert & March (1963) studied collection of individuals at the firm level. They observed rather than seeking maximisation, satisficing is chosen (Simon, 1947; Simon, 1955; Cyert & March, 1963). Decision-makers will seek an alternative to optimal solution and make their decisions once one of the alternative solutions meet a satisfactory condition. Which level the suboptimal solution needs to reach in order to be qualified as satisfactory is determined by decision-makers. Cyert & March (1963) argued individual decision-makers behaviour in searching for alternatives was determined by a trade-off between performance and aspiration level. Performance covers past performance

(13)

but also colleague’s performance while aspiration level is defined as a performance level an individual decision-maker wants to reach (Cyert & March, 1963; Jordan & Audia, 2012). Then, two situations are possible: performance is above aspiration level or performance is below aspiration level. Greve (2003) formulated three implications relating those notions. If performance is above aspiration level, individual search for alternative solutions is decreased, individual likeliness for implementing change is decreased and individual likeliness to choose a solution from riskier ones is decreased. Inversely, if performance is below aspiration level, individual search for alternative solutions is increased, individual likeliness for implementing change is increased and individual likeliness to choose a solution from riskier ones is increased (Greve, 2003; Jordan & Audia, 2012). Jordan & Audia (2012) went beyond the “problem-solving mode” towards the “self-enhancing

mode”. Whereas “problem-solving mode” individuals aim in the first instance at solving problems

while having the same criteria for performance assessment, “self-enhancing mode” individuals estimate their performance as satisfactory while having shifting priority about performance goals and standards over time. This distinction between the two modes is crucial and deepen the extent to which individuals react or don’t react to negative performance feedback. As individuals value a performance level differently whether they are in a “problem-solving mode” or “self-enhancing mode”, individuals appertaining to the last mode would assess a low performance as satisfactory instead of searching for alternative solutions such as “problem-solving mode” individuals. Constructing those two modes, Jordan & Audia (2012) relax Greve (2003) assumptions. Jordan & Audia (2012) add a condition for Greve (2003) assumptions to work when performance is below aspiration level. If performance is below aspiration level, increase in individual search for alternative solutions, likeliness for implementing change and likeliness to choose a solution from riskier one’s assumption will only work if the individual making the decision is in “problem-solving mode”. If the individual rather is in “self-enhancing mode”, he will not assess performance as below aspiration level but satisfactory. Thus, expectations derived by Greve (2003) only possess limited validity.

Based on this literature, we communicated negative prospective financial information that would probably highly threaten municipal councillors’ performance within their municipality. However, given the context of our study, individual decision-makers reaction or ignorance to negative financial information is even more complex. First, as explained by Jordan & Audia (2012), even when faced with negative performance feedback, “self-enhancing mode” individuals tend to ignore them. Similarly, we can expect municipal councillors to lower the standards they considered at the time of our analysis, retrospectively. For our study, it means “self-enhancing mode” individuals would tend to underestimate or ignore the consequences of negative financial information for municipal councillors’ performance. However, it also means “problem-solving mode” individuals would tend to estimate even more critically the consequences of negative financial information. The implication here is that we expect those “two modes” (Jordan & Audia,

(14)

2002) individuals to offset each other as randomization is likely to split them quite equally within the control or treatment group. If we don’t neglect a potentially very little but significant effect of individuals “mode”, we decided to give attention to other variables of interest for which we can expect bigger significant effects but also greater added value to the literature.

At the level of local government politicians, Nielsen & Moynihan (2017) conducted a survey-experiment about Danish officials’ response to performance data. Similarly, as “problem-solving mode” individuals (Jordan & Audia, 2002), they demonstrated local officials give more attention to low performance (Nielsen & Moynihan, 2017). Then, all authors argue providing individual decision-makers with negative performance information is likely to enact a response from them, except for “self-enhancing mode” individuals characterized by Jordan & Audia (2002). Nielsen & Moynihan (2017) expands literature available on behavioural public administration. This study field has been developed and explained by Grimmelikhuijsen, Jilke, Olsen & Tummers (2017). It locates at the intersection between public administration and psychology emphasising how some processes can influence decision-making.

Finally, facing financial decline, van der Voet (2019) provided a useful theoretical framework highlighting organisations could orientate towards decreased or increased innovation as individual decision-makers are looking for solutions to deal with less resources. The main link with our study about van der Voet (2019) work is that the author showed organisational decline as fewer resources for an organisation could lead to more innovative and strategic solutions or not. The extent to which innovation occurs or not is related to environmental context. Also, what can be considered as a new practice varies across countries and organisations as they have different ways of working. Van der Voet (2019) encourages further research at the level of municipalities, notably.

However, we still think it has an influence on directing state preference because Jordan & Audia (2012) demonstrated all individuals did not react the same way to negative financial information then leading to difference in preference for a directing state. Whether individuals are reasoning according to one of the two modes developed by Jordan & Audia (2012), behavioral response might be different. If we have a lot of self-enhancing mode individuals, they are likely to be less influenced by negative financial information treatment then observing less discrepancy between control and treatment groups preference for a directing state. Those individuals would be likely to assess their performance or action as satisfactory lowering the probability of innovation, search for new solutions or shift in policy approach. We acknowledge speaking about aspiration level mean retrospective evaluation with individual looking back as past performances. Then, we would like to clarify some elements. Treatment we give about financial negative information is based on previous economic outcomes that occurred as a result of 2008 economic crisis. Though, it is not incompatible with prospective aspiration levels. Economic crises of 2008 and 2020 are of different

(15)

natures but involves the same mechanisms but in different contexts. With regard to this, their current aspiration levels can be conceived as future performance they try to achieve when knowing about how harmful the effects of an economic crisis are.

2.1.2. Policy positions

Our second central concept is about policy position preferences. We based on Lodge & Hood (2012). Authors work constitutes a solid and relevant theoretical framework for our study. Primarily, they focused on OECD countries from which the countries we ask municipal councillors to participate all are members. Also, they conducted their analysis after 2008 and the financial crisis, meaning they make observations in the context of economic crisis that followed. We also conduct our study in a period of incoming economic crisis, that already started but for which effects could be expected to last for at least several years (Goodman, 2020).

The main contribution we use is about policy response. Indeed, Lodge & Hood (2012) explained states response to declining financial resources was not straightforward. For instance, according to the authors, at least four policy approaches are possible for a state when facing austerity, here Public Service Bargains (PBs). The rationale for a different approach across countries is that it is dependent on institutional settings. Also, the response can be located not only at state level but other levels of government such as municipality and local government ones. Those policy preferences are a directing, hollow, communitarian and coping state (Lodge & Hood, 2012). The first approach, a directing state implies the state as managing a wide range of policies, organising economic policies towards rescue plans in the recovery phases. An important point concerns the lasting specificity of this directing state. It does not only limit to economic interventionism but continues to play a “market-maker” role after economic recovery post-crisis (Lodge & Hood, 2012). Such an approach goes contrary to New Public Management (NPM) and the numerous privatisations occurring in the 1970s and 1980s that have been seriously challenged regarding the financial crisis of 2008.

The second approach, a “hollow state”, bases on outsourcing public services to private firms that negotiate contracts for taking over public services provision. Such an approach is a plausible policy response as last decade financial crisis put even more pressure for cuts and reform on public service. Would a contract be broken before its term by the state or any level of government which contracted it, associated costs would be concentrated on the low number of remaining civil servants and tasks performed by public service (Lodge & Hood, 2012). Then, pressure for cuts would be even more important. By making profit opportunities for private companies, there is only a light intervention of the government in the economy. Government has limited or no control on the process and quality of public service provision.

(16)

The third approach, a “local communitarian state”, confers key responsibilities to community and voluntary organisations (CVOs). While it also involves contract between government and CVOs, those groups are not seeking profit maximisation. Promoting charity principles, they can take responsibility and provide public services in security, self-help and assistance, alternative schooling or disabled individuals insertion, for example. Under this approach, the scope of CVOs is broadened and does not limit to some isolated actions.

The fourth approach, a “coping state”, deals with shortening public service provision. In face of less financial resources and budget cuts, it is also a possibility to provide fewer public services. Only essential and vital public services are maintained while other ones are reduced in number offered to individuals. By reducing the number of civil servants, the quality of public services is also negatively impacted.

2.1.3. Risk-taking

Our third central concept is about risk-assessment. It can be defined as an individual characteristic, the extent to which individuals value their likeliness to take risks and take risks in their daily life, when having to make a decision. According to Frey & all (2017), risk-taking is an individual characteristic that is stable over time. However, a directing state can be perceived as risky for liberals and advocates of a light-touch intervention in the economy because of reduced cost-efficiency and state control.

2.1.4. Debt preferences

Our last final concept involves individuals’ preferences in terms of debt. Public debt defines the amount of money a state, often including lower levels of government, has borrowed and needs to refund loaners with. The relevance of public debt regarding our study lies in the major implications for municipalities. Debt creation capacity or not restrains the scope of public intervention and financing of investments and projects for a municipality. Providing municipal councillors with negative financial information, preferences for public debt have to be taken into account when observing their policy approach preferences.

Increase or decrease in public debt has been argued to be a consequence of the effects of electoral pressure and polarization (Melki & Pickering, 2014; Crivelli, Gupta, Mulas-Granados & Correa-Caro, 2016). Political polarisation could be defined as the discrepancy between political party ideologies within a political system. Last decades, we observed an increase in polarisation in most of European countries as revealed by elections (Pisani-Ferry, 2016). Empirically, Ladner (2014) demonstrated an overall highest polarization within the EU. To conduct his study, he proposed to construct a dimension going beyond distinguishing left and right wings. He included seven policy dimensions

(17)

being “economic liberalization, restrictive financial policy, law and order, restrictive immigration

policy, more environmental protection, expanded welfare state and liberal society” (Ladner, 2014:

5). Regarding the countries we are interested in and their politicisation, Switzerland is at the top (1st_{), France (2}nd_{), Spain (4}th_{on left-right dimension or 5}th_{on seven policies dimension), Netherlands}

(8th_{on left-right dimension or 10}th_{on seven policies dimension) while England is only 24}Th_{or 28}th

(similarly, whether we refer to a left-right or seven policies dimension).

In presence of polarisation, authors views on whether this phenomenon is likely to result in debt increase or decrease diverge.

Melki & Pickering (2014) argued ideology polarisation allow governments to be more resilient when facing demands for an increased debt. They extracted party ideology by reviewing political programmes preferences for public goods and tested the effect between ideological polarization and debt in OECD countries. They concluded there is a negative and significant relation between polarisation and government debt. Thus, their results oppose to Alesina & Tabellini (1990) and Persson & Svensson (1989) whom findings whose support “strategic debt”. Alesina & Tabellini (1990) developed a theory about public debt. They argued public debt was a way for government to restrict future governments set of possibilities they can choose from. They concluded polarisation results in higher public debt because of disagreements between policymakers.

Persson & Svensson (1989) studied the importance of inconsistent preferences, due elections and changes in government, on public debt. A government with embedded ideology towards low public spending like conservatives would not fit with preferences level for a same good for public debt of liberals that would succeed it. More, if a conservative government expects a liberal one to take over the office, it would be more likely to increase its public debt than if it knew it would govern after the end of the actual office. They concluded polarisation leads to public debt.

Then, debt preferences have an influence on the link between negative financial information and policy preferences in the sense some of them are likely to be towards higher debt than others.

2.2. Relationships between variables and hypotheses

This subchapter will present relationships between our four variables and the derived hypotheses.

2.2.1. Relationship between negative financial information and directing state preference This study has for first objective to test the relationship between independent variable being financial decline meaning negative performance feedback and dependent variable being preference for the policy position directing state (Lodge & Hood, 2012). Then, we expect financial decline to have an effect on preference for a directing state. When facing negative financial information, opting

(18)

for a directing state or expressing an increased preference would be explained by the capacity of making officials able to cope with the multiple nature of problems they face. Being competent in almost all sectors allow them to take global actions and make consistent overall plans for economic recovery from a general framework at state level. It would describe what has been qualified of a state comeback by Lodge & Hood (2012).

Hypothesis 1: “Negative financial information positively affects decision-makers’ preference for a directing state approach”.

2.2.2. Influence of risk-taking moderator on the causal relationship between negative financial information and negative financial information.

We argue relationship between financial decline and preference for a directing state is moderated by risk-taking. Indeed, we expect individuals with lower risk-taking to prefer a directing state approach (Lodge & Hood, 2012). The rationale beyond is that a lasting directing state approach nears an emergency response post-crisis but on the long term as explained by Lodge & Hood (2012). A state direction of almost all policy sectors and delegation to municipalities would probably create feelings of security on the individuals. Public provision of services by public authorities refers to the state and his decentralised organisations have a “welfare mission” to ensure the well-being of its citizens. Thus, we expect increase in risk-taking when results in a negative causal relationship between independent and dependent variable

Hypothesis 2: “Negative financial information moderated by risk-taking negatively affects

decision-makers’ preference for a directing state approach”

2.2.3. Influence of debt increase preference moderator on the causal relationship between negative financial information and negative financial information

We argue relationship between financial decline and preference for a directing state moderated by debt increase preference. In accordance with arguments and findings from Alesina & Tabellini (1990) and Persson & Svensson (1989), we expect to observe debt increase preference increases influences positively the relationship between negative financial information and preference for a directing state. An increase in debt is a strategic way to restrain spending of the future government also at local level. More, increased debt means loans and money that may be invested and needed to manage a wide portfolio of competences for municipal councillors.

(19)

Hypothesis 3: “Negative financial information moderated by debt increase preference positively affects decision-makers’ preference for a directing state approach”

(20)

3. Design, data and measurement

In this part, we will explain the design of our study meaning survey-experiment, Then, we will present our data collection strategy. Finally, we will discuss measurement of variables.

3.1. Design

We conducted our analysis at the level of individual council members from local government or municipality, being our unit of analysis, in five European countries: France, the Netherlands, Spain, Switzerland and United Kingdom. As a starting point, all those countries are members of the Organisation for Economic Co-operation and Development (OECD). To make this experiment, we also had to take into account the speaking and knowledge competences of our fellow colleagues leading to this choice of countries to include. Furthermore, increasing the number of individuals we sent out the survey among countries increases the size of our sample. Including more states then more respondents enhances statistical significance of results as demonstrated by the law of large numbers and reduce. Still, we control for differences in countries, first by constructing an interaction term between the independent variable negative financial information and respondent’s country to see difference in preference for a directing state among countries. Then, second, we control for it by observing the effect of respondents’ country as an independent variable and its effect on dependent variable preference for a directing state.

We decided to use a randomised controlled trial by designing a survey-experiment. We think it is particularly interesting to rely on survey-experiment regarding the scope of our study towards behavioural public administration but also psychological effects of the pandemic on individuals (Khan & al, 2020)

The experiment consisted of a survey that first gave information to the respondents about who we are and what is our motive to conduct this study. From the moment they clicked on the link to the survey, respondents were randomly assigned to the control or treatment group. While the control text was about overall information regarding coronavirus, treatment text was composed of around ten lines about negative financial information. Respondents were invited to express their views before having to read about different policy approaches. Then, council members who participated rated their preference for a directing state and other measures linked to variables of interest used by our colleagues for their own studies. Finally, they were asked to fill the survey with some information about themselves such as age, gender, political ideology, risk-taking and debt increase preference to control for manipulation of the experiment and effective randomization. Participants were 2,029 council members split between control group with 963 respondents and treatment

(21)

group with 1,066 respondents. Those 2,029 respondents were from five countries: England for 211 respondents, France for 20 respondents, Netherlands for 1,316 respondents, Spain for 224 respondents and Switzerland for 258 respondents.

.

3.2. Data

Data collection can be characterised as highlighting differences in data availability and access through countries we were interested in. In all countries except for France, it was possible to reach every councillor because they almost all have their own email address publicly accessible. Then, we had to take into account feasibility motives for our sample. For example, in the Netherlands, there are only 355 municipalities in the country (CBS, 2020) so each municipal councillor of each municipality could be asked to answer the survey because of the low amount of municipalities. In comparison, there are 34,970 municipalities in France (Vie Publique, 2019) and the main difficulty lies in the absence of personal email addresses for each municipal councillor. Only the municipality has a public email address email accessible through official published data. To have a similar sample and balance between countries and feasibility reasons, we decided to select only municipalities with more than 10,000 habitants to ask their councillors to fill the survey, excluding overseas departments. We decided to not take them into account because their special socioeconomic and financial situation (Cour des comptes, 2011) would make them be outliers. Still, it concerned around 30,000 municipal councillors. We voluntary included a larger number of potential respondents for France because we could not reach them personally directly. We had to ask for the general services or secretary of the municipality to transmit the email to each of the municipal councillors. We expected then a lower response rate than in countries where it was possible to directly contact them.

3.3. Measurement

We had to develop a measurement strategy for our variables of interest, independent, dependent and the two moderating variables. Our independent variable about financial decline by providing negative performance feedback concerned the treatment group as control group did not receive this information but only general information about coronavirus health consequences. Treatment text also contained a priming effect. Control group only had as displayed a text about people infected or dead because of coronavirus, people who represent a higher risk and expected rising negative effects of the pandemic on mental health. Treatment group only had a text including predicted decreasing tax revenues, rising expenditures concerning social protection and risks of

(22)

unemployment and debt for municipalities. To operationalise whether the respondent was treated or not, we constructed a dummy variable. Indeed, we randomly divided our sample between a control and treatment group: only the treatment group received negative financial prospective information leading to low performance. Control group just was exposed to overall information about health outcomes of coronavirus pandemic. By having two identical groups in terms of individual characteristics composition, excepted the extent to which the treatment is given only to the treatment group, we can derive that differences in the dependent variable are only caused by the treatment. We cannot exclude the threat of having control groups respondents receiving indirectly the treatment. It is possible that fellow respondents communicated about the survey and find out they read different texts when filling their answers. However, we can imagine this effect would be insignificant in a large-n analysis.

If the respondent is treated, the value of the dummy variable associated to his treatment condition is equal to 1. If the respondent is not treated, the value of the dummy variable is equal to 0.

Our dependent variable, preference for directing state measures the preference for an increased reliance on service provision by the municipality after displaying to all respondents a short text about the approach based on Lodge & Hood (2012). We made the text clear, concise and easily understandable. Respondents were asked to rate their preferences for each of the four policy positions described by Lodge & Hood (2012), preceded by a corresponding text. Respondents had to distribute a total of 100 points over the four approaches with the more points are attributed to a position, the more it is preferred. We thought it was more relevant to ask them for a rate than a ranking. A ranking would be used if we wanted to compare preference for the different approaches. However, we only focus on directing state as dependent variable. More, a ranking can have a double signification: it can show which positions respondents prefer or which positions they less dislike without having any absolute preference for any of them. For example, an individual can rank as first an approach while he would only have allocated 30 points out of 100. With ranking, we cannot observe whether the policy position ranked first really involves a preference for the individual. The preference level can also differ widely whether 25 or 100 points are distributed to the first-ranked policy approach. Also, having directly a scale of points makes data management easier because cardinal numbers do not have to be converted in an absolute numerical value, contrary to ordinal values.

Our first moderating variable, risk-taking measures the extent to which council members take risks in their daily life. Here, the same conceptual reasoning has been applied regarding rating rather than ranking as for the dependent variable. However, two main differences are important. First, we opted for a different scale on 10 points rather than 100 points with the more points are attributed the more the respondent is a risk-taker.

(23)

Our second moderating variable, debt increase preference required to ask respondents to what extent the state must increase its public debt on a scale from 0 to 10 with 0 meaning a very low debt increase score and 10 a very high debt increase score. Similarly, as for “M1 second”, we included this measure as part of individual characteristics as preferences for debt vary across individuals.

Finally, to control for country respondents were located in, we created five dummy variables for the five countries they came from. Variable England was coded as 1 if the respondent was located in England, 0 otherwise. Variable France was coded as 1 if the respondent was located in France, 0 otherwise. Variable Netherlands was coded as 1 if the respondent was located in Netherlands, 0 otherwise. Variable Spain was coded as 1 if the respondent was located in Spain, 0 otherwise. Variable Switzerland was coded as 1 if the respondent was located in Switzerland, 0 otherwise.

(24)

4. Empirical findings

This chapter is organized in four different subsections. We first performed tests for normality for our data to check if the student’s t-test required assumption of normality is respected. Then, we could do manipulation checks and descriptive statistics. Finally, we regressed the effect of negative financial information dependent variable on preference for a directing state independent variable adding the two moderating variables separately, risk-taking and debt increase preference. Then, we would be able to observe their moderating effects on main causal relationships providing an empirical answer to our research question and regarding our hypotheses. Finally we wanted to observe variation produced by being a respondent from country citizenship within treatment group.

4.1. Test for normality of data

Before using t-statistics to compare similarity between groups characteristics and preferences, we needed to check for normality of the t-distribution. An important step was the removal of outliers signaling extreme values in our data. Usually, such procedure can be controversial when the rationale for dropping those numbers is motivated on conceptual grounds. In our study, we removed some values considered as outliers because it would not make sense to keep them for analysis as those values deviate clearly from reality. We removed outliers for respondents age and time spent reading about the text displayed about policy preference. Indeed, we removed 78 observations when describing age because of values in years over 120 and 2 for values under 18 probably due typing mistakes. Knowing that older individual on earth does not exceed 120 years of age and being 18 years old is a legal requirement to be a municipal councilor, those extreme values would have distorted our results. Now, removing outliers for time needs a bit more justification. By time spent reading the text, we measure time spent between directing state approach text displaying and directing state approach text page submission to continue the survey. To observe the results for this indicator, we made the choice to remove 58 observations with a duration less than 1 second. For sure, it seems possible to spend less than 1 second to go to the other page. Though, we highly doubt those people are able to read around 10 lines so fast. However, it does not mean they did not know what the approach was about. As we noticed them they will have further description about the policy positions in the next pages, we can imagine a description of a few words was enough for them to understand the concept of the directing state approach. Also, we removed 2 observations with a duration over 300 seconds. Explanation for those values might be people who did not complete the survey straightforwardly or were interrupted when doing it and came back to it a few times after while time spent measure was running. Asking municipal councillors, literacy, numeracy or understanding issues is not likely to happen and 5 minutes might

(25)

be enough even so. Note that for both age and time, we are dealing with outliers for variance tests and descriptive statistics that would have increased the variance and the mean and not outliers for regression analysis that would be treated in subchapter 4.3.

Thus, we tested distribution normality for time, directing state preference score, gender, age, council position, ideology, risk-taking and debt preference. We used Shapiro-Wilk W test for normality for 4≤N≤2000 and Shapiro-Francia W’ for 5≤N≤5000. We used a significance level for p of 5%.

Table 1

Tests for equality of variance between treatment and control groups for respondents’ characteristics and preference

Note: Significance level of 5%.

Table 1 shows all variables except gender have a p<0.001 while a p<0.05 is sufficient to reject the null hypothesis of equal variances between treatment and control groups. This comparison was necessary to know which type of t-test we have to perform for descriptive statistics. Consequently, we face unequal variances while having removed outliers. Implication is that we will perform Welch’s t-tests because it does not require assumption of equal variances and we demonstrated variances were not equal for almost all variables. Student’s t-test will only be used for gender because p=0.0950 and a p>0.05 is sufficient to fail to reject the null hypothesis, meaning no rejection of equal variances.

4.2. Manipulation checks and descriptive statistics

Randomized controlled trials (RCTs) are very powerful in explaining causal relationships but require specific precautions when analyzing results (Angrist & Pischke, 2015). Indeed, every variation between the control and treatment groups is explained by the condition respondents received the treatment or not as long as randomization worked efficiently and correctly. Thus, it is

(26)

of primary importance to check whether groups composition is similar in terms of respondents’ characteristics such as age or gender. For example, a significant variation in age between groups could be responsible for an underlying variation in preference for a directing state making the treatment unable to be sufficient itself in explaining the difference in groups outcome while still being necessary though.

Regarding the following tables, we test if we reject or fail to reject the null hypothesis. The null hypothesis, meaning there is not significant statistical differences between groups cannot be rejected if p>0.05 with a significance level of 5%. The alternative hypothesis, meaning there is significant statistical differences between groups implies rejection of the null hypothesis if p<0.05 with a significance level of 5%. A large t>2 or t<-2 gives evidence for rejection of the null hypothesis of no treatment effect. Reversely, a small -2<t<2 means a failure to reject the null hypothesis. With descriptive statistics, we wanted to observe whether there is significant variation in respondents’ characteristics or not. Such a variation in those respondents’ characteristics could influence the scoring for a directing state preference leading to a misinterpretation of the true effect of the treatment.

Table 2

Descriptive statistics for individual characteristics.

Note: Welch’s t-tests and Student’s t-test for individual characteristics of respondents. Standard errors are in parenthesis. Significance level of 5%.

We included time respondents spent reading the text we wrote for respondents about what a directing state approach would look like in application. We chose to include this measure because it is a good indicator of how well understood was the policy approach description.

From Table 2, we see treatment group respondents took on average around two more seconds to read the text displayed to them about the policy approach of interest. However, tvalue of -1,0071 is not significant in rejecting the null hypothesis then stating this difference in means between groups is statistically insignificant. Concerning p-value, being superior to 0,05, we can draw the conclusion group assignment did not lead to significant difference in time spent reading about the text, at least not at any level strictly superior to 69%. Then, effect from independent to dependent variable is produced by the content of the treatment, not the length or format.

(27)

Similarly, we performed such t-statistics to check whether groups are similar in respondents’ characteristics composition. Indeed, every potential significant variation between the control and treatment groups would be explained by the condition respondents received the treatment or not as long as there is no significant difference in groups’ characteristics. Thus, it is of primary importance to check whether groups composition is similar in terms of respondents’ characteristics such as age or gender. For example, a significant variation in age between groups could reveal a randomization that did not work well.

From Table 2, we can deduct both groups contained more male than female as both coefficients around 0.29 are nearest to 0 than 1. Having a t-value of 0.0968 indicates we fail to reject the null hypothesis, saying the very low variation in means between groups is insignificant. High p=0.9229 is largely superior the significance level of 5%. Then, we are able to state this very low insignificant in means is not due to treating part of respondents. Indeed, it seems quite logic no causal relationship exists between treatment condition and gender in case of effective randomization.

Table 2 illustrates groups means for age are almost similar around 55 years old. As for gender, t is within -2 and 2 then being insignificant statistically rejecting the alternative hypothesis of difference between means of groups. p=0.4572>0.05, we can say chance explains this slight insignificant difference at any level not higher than 45,72%.

Table 2 reveals in both groups, means are nearest 1 than 0 then municipal councillors were mostly from majority than minority even though mayors slightly strengthens this coefficient. This very little variation between means of groups is statistically insignificant with t=0.5510 and due to chance with p=0.5817.

Table 2 shows means nearest the centre while being though more towards left-wing as being under 5. In other words, respondents are from the centre with little preference for left-centre than right-centre. As visible when having a look at means, variation between groups is highly insignificant with t almost equal to 0. Those results are the product of chance with p=0.9994.

Table 2 reveals little variation between groups means that both indicate respondents are moderate risk-takers with slight preference for risk-taking as the mean is nearest 10 than 0 over being risk averse. Little difference in means is highly insignificant with t=-0,0258. Regarding p=0.9795, we fail to reject the null hypothesis of differences due to chance rather than experimental design.

(28)

Table 2 reveals little variation in means of debt increase preference scoring. Note that we asked respondents preference for debt increase and not just for debt. Even a low score signals a preference for debt increase while 0 is the only possible value for an absolute non-preference for debt increase. On average, respondents in both groups attributed a score nearest 10 than 0 and nearest 6 than 5 meaning they prefer a consistent debt increase. Difference in means is insignificant as t=-0.4672 and due to chance because of failure to reject the null hypothesis.

Table above demonstrated randomization worked correctly: no significant difference explained by the treatment have been revealed. We know individuals with similar characteristics and preferences for risk-taking and debt increase preference attributed different scores for the policy approach.

4.3. Regression of negative financial information on directing state preference and

influence of moderating variables risk-taking and debt increase preference.

This subchapter is divided in three parts: effect of independent variable on dependent variable, effect of risk-taking moderating variable on the causal relationship between independent and dependent variable and effect of debt increase preference moderating variable on the causal relationship between independent and dependent variable.

4.3.1. Effect of independent variable negative financial information on dependent variable directing state preference.

As we demonstrated significant differences in policy approach scoring between groups due to treatment, we can now perform a linear regression to observe the effect of negative financial information on preference for a directing state. The linear regression follows the equation Y=a+b*X+e with intercept a, b*X being the treatment coefficient b multiplied by the treatment or independent variable “X” and e for residuals.

(29)

Table 3

Linear regression table between negative financial information and preference for a directing state

Note: Standard errors are in parenthesis. Level of significance of 5%.

Table 3 summarizes main findings when regressing the effect of negative financial information on directing state preference. We observe treating part of respondents by providing them with negative financial information led to a score decrease of around 4.5 points when expressing their preference for the policy approach we are interested in. Compared to the control group, treatment group attributed a score lower by around 4.53% as scoring scale ranged from 0 to 100 then points being equal to percent. With a 95% certainty, we know this coefficient varies across around [-7;-2] for the 2,029 respondents. As t=-3.60, this coefficient is significant and as p<0.001, we can say with a strong level of confidence, this coefficient is due to experimental treatment and not chance. Regarding the first hypothesis we formulated, “Negative financial information positively affects

decision-makers’ preference for a directing state approach” (subchapter 2.2.1, p.14), the results are

not the ones we were expecting. The effect is significant, but negative? Indeed, there is negative causal relationship between the independent variable and the dependent variable meaning an increase in the categorical independent variable from 0 to 1 leads to lower scores for directing state preference. Also, a negative effect of 4.53% cannot be neglected. Finally, R-squared indicates that the independent treatment variable explains only 0.63% of the variance in the dependent variable. To know which scores groups have attributed for a directing state, we can perform a Welch’s t-test. It also shows with t-value and p-value that variation is significant and due to experimental.

(30)

Table 4

Welch’s t-test for directing state preference scores for control and treatment groups.

Note: Standard errors are in parenthesis. Significance level of 5%.

As showed in table 3, we observe a difference of around -4,5% when comparing directing state preference scores attributed in control and treatment groups from 41,96 points to 37,42 points.

Figure 2

Note: Graph representing respondents scores for directing state preference in control and treatment group.

Figure 2 illustrates respondents score attributed to a directing state approach in control and treatment group. It looks like a lot of points are located on the horizontal axis for a preference equal to 0 points in both cases. We see fitted regression line by the method of ordinary least squares is

(31)

lower on the vertical axis for the treatment group than for control group as treatment group had less preference for a directing state. Indeed, it is also visible with figure 3.

Figure 3

Note: Graph representing real data distribution for treatment and control groups compared to normal distribution.

As demonstrated numerically by performing Shapiro-Wilk and Shapiro-Francia tests for equality of variance, figure 3 help us to observe density for different score values for directing state preference. We note a high density for a score of 0 in both groups and particularly for treatment group. In both cases, we graphically see some skewness on the left part of the graphs, even more skewness to the left is observable for treatment group. Also, we see some kurtosis on both graphs, meaning the two tails outside the 95% confidence interval have unequal weight. Those statistics illustrate the departures from normal data more understandable than just by performing statistical tests.

(32)

4.3.2. Effect of moderating variable risk-taking on the causal relationship between negative financial information and directing state preference.

We wanted to observe the effect of risk-taking on causal effect between negative financial information and directing state preference. We test risk-taking as a moderator first by creating an interaction term between negative financial information and risk-taking. Then, we test risk-taking as an independent variable.

Table 5

Regression table of risk-taking influence as a moderator on relationship between negative financial information and directing state preference. Standard errors are in parenthesis. Level of significance of 5%.

Note: Standard errors are in parenthesis. Level of significance of 5%.

Table 5 describes results for control and treatment groups. We see moderated main effect is still negative, but higher than without taking into account risk-taking (table 3).

Regarding treatment group, we have negatively and considerable significant p-value and t-value. Concerning hypothesis 2: “Negative financial information moderated by risk-taking negatively affects decision-makers’ preference for a directing state approach” (subchapter 2.2.2, p.15), we cannot reject our hypothesis as treated respondents attributed -1,25% scoring preference for a directing state. We know this effect is due to treatment (p<0.001) and reveal significant variation within treatment group (t=-3.76). Adjusted R-squared of 0,0076% means adding risk-taking to the model gives slightly more explanatory power for directing state preference while still being very low.

Strategy of European local municipalities officials to deal with financial decline in the context of the Covid-19 pandemic: An analysis of the influence of risk-taking and debt preferences by survey experiment